Remote sensing image analysis by aggregation of segmentation-classification collaborative agents. (January 2018)

• Record Type: Journal Article
• Authors: Troya-Galvis, Andrés; Gançarski, Pierre; Berti-Équille, Laure
• Is Part Of: Pattern recognition. Volume 73(2018:Jan.)
• Page Start: 259
• Page End: 274
• Publication Date: 2018-01
• Subjects: Segmentation -- Classification -- Collaborative approaches -- Remote sensing; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.08.030
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Two new approaches for collaborative remote sensing image analysis are presented. They both are based on a multi-paradigm framework which uses classification to guide a segmentation process. The proposed methods aggregate many mono-class extractors in order to make multi-class remote sensing image classification. Experiments show that the proposed methods give better results (both in terms of classification and segmentation) than a hybrid object-based approach as well as a deep learning approach, even if the training data is limited in quantity and quality. Abstract: In this article we present two different approaches for automatic remote sensing image interpretation which are based on a multi-paradigm collaborative framework which uses classification in order to guide the segmentation process. The first approach applies sequentially many one-vs-all class extractors in a manner inspired by cascading techniques in machine learning. The second approach applies many collaborating one-vs-all class extractors in parallel. We show that the collaboration of the segmentation and classification paradigms result in a remarkable reduction of segmentation errors but also in better object classification in comparison to a hybrid pixel-object approach as well as a deep learning approach.